Questions customers ask before buying, and where each answer belongs

Search this phrase and the first page gives you numbered lists. Five questions, seven questions, fifty-two questions. They are written from the seller's chair, from memory of what buyers usually want to know, and the rows are not wrong: price, risk, proof, comparison, fit. The trouble is that no row carries a state. The list tells you what gets asked. It says nothing about what happened when somebody asked it.

Outcome is the expensive part. In the 2025 B2B Buyer Experience Report from 6sense, built on roughly four thousand responses plus a companion study of 766, 94% of buying groups put their shortlist in order of preference before contacting any seller, and the top-ranked vendor won about 80% of the time. The same study found buyers reaching out six to seven weeks earlier than the year before, and one of the reasons was blunt: 62% needed a conversation with a seller because questions about capabilities, cost and security were not answerable from what the vendor had published. A question that reaches a person on a call has usually already gone looking for an answer somewhere else and come back empty.

Consumer data shows the same shape at a volume nobody has on calls. alby's read of shopper questions covers more than 4.6 million questions from more than 1.8 million on-site interactions between September 2024 and September 2025. One question dominates: what other customers say about the thing. It appeared more than 121,000 times, roughly four times as often as the next most common one. Fit, styling and wear accounted for about 18% of everything asked, close to 845,000 questions. Those are questions a page can settle in a sentence, and when the page does not, they turn up later in an inbox or on a call.

The published lists are sorted by topic, and topic is the least useful axis

Sorting buyer questions by topic produces a reference document. Sorting them by what happened produces a work queue. Topic tells you that cost comes up often, which you already knew. Outcome tells you which of those cost conversations ended with the buyer still not knowing, and that smaller set names a handful of specific conversations and a piece of writing that does not exist yet.

Most research on questions in sales conversations runs in the other direction. Bain's November 2025 read of sales call transcripts at a health tech company is about the questions the seller asks: discovery depth, consistency, staying in the conversation long enough to win later. That research is useful and it answers a different question. The buyer's own questions are a separate corpus, and they are the one that tells you what the market could not find out on its own.

So the first thing to do with a month of recordings is not to count topics. It is to attach an outcome to every question that was asked out loud, and to accept that three of the four outcomes are interesting. Nobody has time to do that pass by hand, which is why the topic list survives. An hour of your own recordings is enough to see which outcome dominates.

A question nobody answered and a question answered badly need different owners

Four things can happen to a question on a call, and they are not degrees of the same thing. The question can be answered properly. It can be answered, badly. It can be asked and left hanging. Or the buyer never asked, which is also information when the topic was obviously live. Each outcome points at a different repair.

The fourth row is the one teams skip, and it is worth keeping. When a topic was obviously live in the conversation and the buyer never raised it, that silence is usually not comfort. It means the buyer had already settled the matter somewhere else, with a review site, a peer, or a competitor's page, and whatever they settled on is now fixed and invisible. A topic that disappears from questions while the deals keep stalling is a signal to chase rather than a box that got ticked.

The second and third rows look identical in a CRM note, because a note records what was decided and not what was asked. They cost different things. A vague answer is a positioning problem that one person can fix in an afternoon. A question with no answer anywhere is a roadmap conversation. Grouping them under customer questions, as every list does, means the expensive one usually ends up with the cheap remedy.

A note on the seller side of this, because it is a different job: deciding which calls get opened at all, and what a reviewer looks for when they open one, belongs to the call review process rather than to this list. The buyer's questions are what you harvest from calls somebody already reviewed, or from calls nobody will ever have time to review.

Most requests left hanging are not product gaps

When we built the screen that collects requests which went unanswered, the framing we wanted first was the dramatic one: call the block product gaps and let the number do the work. The flag underneath does not support that reading. It marks a request that got no answer inside the conversation, and most of what it catches is ordinary: documents that were never sent, a callback that did not happen, a second appointment nobody booked, a technical problem on the line.

The same discipline applies to the way a pain gets worded. So the block kept the duller name, requests left unanswered, and in Spellit ClientTruth the real product gap is a subset inside it, the requests carrying the label for a specific product the client wanted and we do not have. The honest framing produces a smaller number. The smaller number is the one a product manager can act on without auditing it first, and the larger one would have been argued away within a week of anybody checking it.

This is the practice worth copying whatever tooling you use. Keep one bucket for every request that died in the conversation, because that bucket is cheap to fill and useful on its own. Then carve out the subset where the answer does not exist in the company at all. The first bucket is an operations problem and it belongs to whoever runs the handoffs. The second is the only one that should ever reach a roadmap. Mixing them is how a product team learns to ignore the whole report: three reviews in, somebody notices that half the product gaps were a missing PDF, and the report stops being opened. Telling a documents problem apart from a real gap is the sorting we would happily do alongside you.

The list of questions is always longer than the list of themes

The second thing we got wrong was the vocabulary. The model that labels client questions receives a list of slugs as examples, and the code behind it validates only the shape of the string: lowercase letters, digits and hyphens, between two and forty characters. It never checks meaning against the list. The model invents labels freely, and every invention lands in the mart as a first-class label.

What came back was roughly two hundred and fifty distinct question labels where about a dozen carried any meaning. Subsidy eligibility sat alongside a dozen near-synonyms of itself. The screen now folds the tail into Other, which is a display fix for an authoring problem. Worth naming what is different here from the usual story of a noisy list: nothing broke, and no filter was missing. The validator did exactly what it was written to do. The noise was authorized earlier, at the moment we decided that a vocabulary could be a hint rather than a rule, and that decision is invisible until you read the output.

The operational consequence lands on anybody building site content from this. A raw label list is not a content plan. Merge first, count after. One page per label gives you forty thin pages that compete with each other; one page per merged theme gives you a dozen that can hold a position. The same arithmetic shows up on the seller's side of the data, where an objection taxonomy grew to 424 labels before anybody asked how many of them described different things.

A question is worth ranking only against the calls where it could have come up

A theme that came up in 40% of the calls where it could plausibly come up is a different finding from a theme that came up in 40% of all calls, and the second number is almost always the one people quote. Base matters more than the percentage. Share of questions is counted against conversations that contained a question at all; share of requests against conversations that contained a request. Counting either against every call in the period hides the themes that only surface in a specific kind of conversation, which are the ones worth writing a page about.

There is a second base worth keeping separate, and it answers a different question. Week-over-week movement is read against every call in the week, because the thing you want to know there is whether a theme is spreading across the whole book, not whether it dominates among the people who happened to ask something. Both charts sit on the same screen with different bases underneath, so each one has to say what it counts against. Without that, the same theme reads as two different numbers and somebody quotes whichever is larger. The same care applies one level down: an item that did not apply to a conversation is not a zero, and counting it as one quietly inflates the base, which is the argument in why a scoring item marked not applicable is not a failure.

For the seller side of the same recordings, where the unit of analysis is what the rep did rather than what the buyer asked, the mechanics live in Spellit Actions. The two readings use the same conversations and almost never produce the same priority list, which is the point of running both.

What to do next

Take twenty recordings from last month, ten won and ten lost, and do one pass with a single column: every question the buyer asked out loud, written in their words. Do not categorize yet. Then do a second pass and put one of four marks against each row: answered, answered badly, asked and dropped, or no answer exists. The second pass is where the work actually is, and it goes faster than the first.

Then split the sheet three ways. Rows marked answered badly go to whoever owns the script and the pages, with the buyer's wording attached, because the wording is the deliverable, and a paraphrase loses the part that was useful. Rows marked asked and dropped go to the prep sheet for that call type. Rows marked no answer exists go into one list with a count, and that list is reviewed monthly rather than discussed in the moment.

One check before you trust the sheet: open your own site and count how many of the twenty questions have an answer a stranger could find without talking to anybody. That count is the honest size of the content gap, and it is the one number here that nobody in the room can argue with, because anybody can repeat the check themselves. Putting an FAQ page next to a month of real buyer questions is where the overlap usually surprises people.